- Email: [email protected]
Predictive values of acute coronary syndrome discharge diagnoses differed in the Danish National Patient Registry
Journal of Clinical Epidemiology 62 (2009) 188e194
Predictive values of acute coronary syndrome discharge diagnoses differed in the Danish National Patient Registry Albert Marni Joensena,*, Majken K. Jensenb, Kim Overvadb, Claus Dethlefsenc, Erik Schmidta, Lars Rasmussena, Anne Tjønnelandd, Søren Johnsenb a
Department of Cardiology, Center for Cardiovascular Research, Aalborg Hospital, Aarhus University Hospital, Sdr. Skovvej 15, DK-9000 Aalborg, Denmark b Department of Clinical Epidemiology, Center for Cardiovascular Research, Aarhus University Hospital, Aalborg, Denmark c Center for Cardiovascular Research, Aalborg Hospital, Aarhus University Hospital, Aalborg, Denmark d Danish Cancer Society, Institute of Epidemiology, Copenhagen, Denmark Accepted 31 March 2008
Abstract Objective: To investigate the predictive value of acute coronary syndrome (ACS) diagnoses, including unstable angina pectoris, myocardial infarction, and cardiac arrest, in the Danish National Patient Registry. Study Design and Setting: We identified all first-time ACS diagnoses in the Danish National Patient Registry among participants in the Danish cohort study ‘‘Diet, Cancer and Health’’ through the end of 2003. We retrieved and reviewed medical records based on current European Society of Cardiology criteria for ACS. Results: We reviewed hospital medical records of 1,577 out of 1,654 patients (95.3%) who had been hospitalized with a first-time ACS diagnosis. The overall positive predictive value for ACS was 65.5% (95% confidence interval [CI] 5 63.1e67.9%). Stratification by subdiagnosis and hospital department produced significantly higher positive predictive values for myocardial infarction diagnoses (81.9%; 95% CI 5 79.5e84.2%) and among patients who received an ACS diagnosis in a ward (80.1%; 95% CI 5 77.7e82.3%). Conclusion: The ACS diagnoses contained in hospital discharge registries should be used with caution. If validation is not possible, restricting analyses to patients with myocardial infarction and/or patients discharged from wards might be a useful alternative. Ó 2008 Elsevier Inc. All rights reserved. Keywords: Validation studies; Coronary disease; Myocardial infarction; Registries; Unstable angina pectoris; Diagnosis
1. Introduction The mortality from coronary heart disease (CHD) has declined over recent decades in most industrialized countries; however, CHD remains a leading cause of death and morbidity [1e4]. Major efforts have been made by physicians, administrators, and politicians to further reduce the risk and further improve the prognosis of CHD. Therefore, it is essential that up-to-date and valid data on the incidence and outcome of CHD are available. Hospital discharge and other disease registries are cost-efficient data sources; however, their usability is highly dependent on the validity of the registered data. Previous studies on the validity of CHD diagnoses contained in hospital discharge and other disease registries
* Corresponding author. Tel.: þ45-99-32-68-39. E-mail address: [email protected] (A.M. Joensen). 0895-4356/09/$ e see front matter Ó 2008 Elsevier Inc. All rights reserved. doi: 10.1016/j.jclinepi.2008.03.005
have focused primarily on myocardial infarction. In general, these studies have reported sensitivities with some variation in the positive predictive value (65e96%) [2,5e9]. Few studies have examined the validity of unstable angina pectoris and cardiac arrest diagnoses, and most of these studies were performed before the introduction of the International Classification of Diseases’ 10th revision (ICD-10). Moreover, the strategy for diagnosing patients with suspected acute coronary syndrome (ACS), that is, myocardial infarction, unstable angina pectoris, and cardiac arrest, has changed considerably during the last few years due to the introduction of sensitive and specific biomarkers of myocardial necrosis and a new definition of myocardial infarction that includes troponin levels [10]. These changes would presumably have implications for the diagnostic workup and validity of the ACS diagnosis. Therefore, we examined the predictive value of myocardial infarction, unstable angina pectoris, and cardiac arrest diagnoses contained in the Danish National Patient Registry for
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
participants in the cohort study ‘‘Diet, Cancer and Health’’ from 1994 to 2003.
2. Methods 2.1. ‘‘Diet, Cancer and Health’’ ‘‘Diet, Cancer and Health’’ is a prospective cohort study with the primary objective of analyzing the etiological role of diet in the development of cancer. The study has been described in detail elsewhere [11]. Between December 1993 and May 1997, 80,996 men and 79,729 women aged 50e64 years were invited to participate in the study; 27,179 men and 29,876 women accepted the invitation. Those born in Denmark, living in the urban areas of Copenhagen and Aarhus, and not registered with a cancer diagnosis in the Danish Cancer Registry at the time of invitation were eligible cohort members.
189
Patient Registry, including primary and secondary diagnoses. Participants who had an ACS discharge diagnosis before enrollment in ‘‘Diet, Cancer and Health’’ were excluded. Only participants who had received a first-time ACS diagnosis in the Danish National Patient Registry after the time of their enrollment in ‘‘Diet, Cancer and Health’’ until January 1, 2004 were included in the study. We noted whether the department of admission was an emergency room, an outpatient clinic, or a ward, and whether the hospital of admission had specialized functions of invasive cardiology or not. If patients who visited an emergency room were admitted to a hospital ward and were registered with an ACS diagnosis by both departments, they were categorized as patients from a ward. When registered with an ACS diagnosis from an emergency room (or an outpatient clinic) without a subsequent ACS diagnosis from a ward, they were categorized as patients from an emergency room (or an outpatient clinic).
2.2. Identification of possible cases of ACS The Danish National Health Service provides free universal tax-supported health care, including hospital care, for all inhabitants (Health Care in Denmark. Copenhagen: Ministry of the Interior and Health, 2003; available at: http://www.im.dk/publikationer/healthcare_in_dk/index.htm). Patients with acute medical conditions are admitted to and treated in public hospitals. The Danish National Patient Registry was established in 1977, and 99.4% of all discharges from somatic hospitals (including public and private hospitals) have been recorded in this database [12]. If a patient was transferred from one department to another, the patient was registered with a discharge diagnosis from the first department at the time of transfer and subsequently registered with a second diagnosis when discharged from the second department. The diagnoses from the two departments need not be identical. Since 1995, discharges from emergency rooms and outpatient clinics have also been included in the registry. The registered data from each admission include the civil registry number, which is the unique personal identification number of every Danish citizen, the dates of admission and discharge, the surgical procedures performed, and up to 20 discharge diagnoses classified according to the Danish version of the International Classification of Diseases, 8th revision (ICD-8) until 1993. Since 1994, diagnoses have been classified according to the corresponding national version of ICD-10 [12]. All discharge diagnoses are determined exclusively by the physician who discharges the patient and cannot be altered later, for example, for administrative or financial reason. Based on the available hospital discharge history of each participant, we identified participants in the ‘‘Diet, Cancer and Health’’ cohort who were registered with a discharge diagnosis of ACS (ICD-8 410e410.99, 427.27 and ICD10 I20.0, I21.0eI21.9, I46.0eI46.9) in the Danish National
2.3. Medical record review We retrieved medical records corresponding to the discharge diagnosis and exact discharge date contained in the National Patient Registry from 54 different hospitals. We sought to retrieve the medical records that covered the hospital admissions recorded in the National Registry of Patients; however, minor mismatches (eg, 1 or 2 days in admission or discharge date) between the registry and the medical records were accepted if all other available information indicated that the medical record covered the admission identified in the registry. The records were reviewed by one of three reviewers (AMJ, SPJ, or MKJ). If the complete medical record was not available, discharge letters, results of blood tests, and electrocardiograms (ECG) were retrieved whenever possible. Information about clinical symptoms, ECGs, and biomarkers of myocardial infarction was noted. If the patient had died within 28 days of developing a myocardial infarction, the date of death was noted and the autopsy record, if any, was retrieved. We also noted the duration of the hospital stay. Patients were classified in accordance with the current recommendations of the American Heart Association and the European Society of Cardiology, as described by Luepker et al. [13]. This classification is based on symptoms, signs, coronary biomarkers (creatine kinase, creatine kinase B, creatine kinase (CK) MB, troponin T, troponin I, lactate dehydrogenase, and lactate dehydrogenase 1), and ECG and/or autopsy findings. All cases with an uncertain diagnosis based on the available information were discussed, and diagnoses were made according to consensus. We used the Civil Registration System, which retains data on vital status, address, and emigration for the Danish population since 1968, to obtain information on patient deaths [14].
190
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
For more details on the classification of ACS events, please see Luepker et al. [13]. 2.4. Statistical analyses Patients with definite myocardial infarction (fatal and nonfatal events), probable myocardial infarction, possible myocardial infarction, unstable angina pectoris, or a medical procedure-related event were categorized as verified ACS. Patients with no unstable coronary events or unrecognized myocardial infarction were categorized as verified non-ACS. Patients for whom information was insufficient to categorize them as either verified ACS or verified nonACS were excluded from the analyses. Positive predictive values of ACS diagnoses registered in the Danish National Patient Registry were calculated as proportions, that is, the numerator contained the number of patients with verified ACS after review of medical records and the denominator contained the total number of patients registered in the Danish National Patient Registry with this specific diagnosis. Positive predictive values were calculated for ACS as a whole and for each specific ACS subdiagnosis (unstable angina pectoris, myocardial infarction, cardiac arrest). Furthermore, data were stratified based on the type of department (ward, emergency room, or outpatient clinic), length of hospital stay, time of diagnosis (before or after January 1, 2000), type of hospital (whether the hospital of admission had specialized functions of invasive cardiology or not), and gender. Finally, we conducted a multivariable logistic regression to estimate the risk of having a validated diagnosis in order to identify characteristics independently associated with the positive predictive value. In this analysis, we included the following factors: subdiagnosis, type of department, gender, age, length of stay, type of biomarkers (troponins/ CKMB, others, none), and type of diagnosis (primary or secondary). All the factors were mutually adjusted.
3. Results We identified 1,654 patients with an incident diagnosis of ACS in the Danish National Patient Registry among the participants in the ‘‘Diet, Cancer and Health’’ study. Baseline characteristics of the cohort are presented in Table 1. The cohort provided 419,949 person-years of risk resulting in an incidence rate of first-time ACS diagnosis in the Danish National Patient Registry of 3.9 events per 1,000 person-years (95% confidence interval [CI] 5 3.7e4.1). We were able to retrieve medical records or discharge letters for 1,577 (95.3%) of the patients. We were not able to characterize 96 patients, either because we could not retrieve their medical records (n 5 77) or because their medical records had insufficient data to classify these patients (n 5 19). From the medical records, we were able to find information on symptoms (n 5 1546; 98.0%), ECG
Table 1 Baseline characteristics of the ‘‘Diet, Cancer and Health Study’’ population Characteristics
Study population
Age at enrollmenta (yr) Current smokers (%) Body mass index O25 kg/m2 (%) Total cholesterola (mmol/L) Blood pressurea (mmHg) Diabetes mellitus (%) Education !8 yr (%)
56.1 (52.4e60.3) 36.1 55.9 6.0 (5.3e6.8) 138 (125e152) 2.0 32.8
a
Median (1. quartile and 3. quartile in brackets).
(n 5 1,501; 95.2%), all biomarkers (n 5 1,363; 86.4%), and highly specific biomarkers, troponins, and CKMB (n 5 1,290; 81.8%). Table 2 shows how patients are classified in accordance with the American Heart Association and European Society of Cardiology guidelines. Classification of the ACS diagnoses is presented in Table 3. The positive predictive value for ACS as a whole after exclusion of those with missing medical records was 65.5% (95% CI 5 63.1e67.9%) (1,021/1,558). After stratification by subdiagnoses, we found a substantially higher overall positive predictive value for myocardial infarction (81.9%; 95% CI 5 79.5e84.2%) (878/1,072) compared to unstable angina pectoris (27.5%; 95% CI 5 23.4e31.9%) (122/444) and cardiac arrest (50.0%; 95% CI 5 34.2e65.8%) (21/42). When we stratified the data based on the type of department of discharge, the predictive value for patients diagnosed in a ward was much higher (80.1%; 95% CI 5 77.7e82.3%) (964/1,204) than that for patients diagnosed in an emergency room or in an outpatient clinic (16.1%; 95% CI 5 12.4e20.4%) (57/354). There were 1,021 patients with a verified diagnosis of ACS, 834 (81.7%) of whom were discharged from a ward with a diagnosis of myocardial infarction. Only 57 (5.6%) of the patients with verified ACS were discharged from an emergency room or from an outpatient clinic (among these, 10 patients died). Nearly all the patients registered with an ACS diagnosis from an emergency room or outpatient clinic were transferred to a hospital ward and subsequently discharged with diagnoses other than ACS. These patients often had stable angina pectoris or had chest pain for other reasons. The positive predictive value was as high as 92.4% (95% CI 5 90.4e94.0%) (834/903) for patients diagnosed with a myocardial infarction in a ward compared to 26.0% (95% CI 5 19.6e33.3%) (44/169) for patients diagnosed with a myocardial infarction in an emergency room or in an outpatient clinic (Table 4). Stratification based on the length of hospital stay (longer vs. 2 days or less) produced similar but not so pronounced differences in positive predictive values for patients with hospital stays longer than 2 days (for all diagnoses, 78.8%; 95% CI 5 76.3e81.2% [883/1,120] and for
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
191
Table 2 Classification of cardiac events based on symptoms/signs, ECGs, and/or biomarkers, as recommended by the American Heart Association and European Society of Cardiology Definite myocardial infarctiona
Diagnostic biomarker findings or evolving ECG Or, if fatal, typical autopsy findings Equivocal biomarker findings and positive ECG If biomarkers are missing, positive ECG and typical symptoms Equivocal biomarker findings and (nonspecific ECG or typical symptoms) If biomarkers are missing, positive ECG Normal biomarkers and positive ECG and typical symptoms Cardiac event (as described above) within a period of 28 days after a medical procedure (eg, general surgery) Appearance, in a nonacute setting, of a new diagnostic Q wave
Probably myocardial infarctiona Possible myocardial infarctiona Unstable angina pectoris Medical procedure-related event Unrecognized myocardial infarction
Abbreviations: ECG, electrocardiograms; CHD, coronary heart disease. Symptoms or signs were categorized as either typical (acute chest, epigastric, neck, jaw or arm pain, discomfort or pressure without any apparent noncardiac source, or acute congestive heart failure in the absence of non-CHD causes) or atypical. ‘‘Evolving ECG’’: evolvement of a Q code. ‘‘Positive ECG’’: evolving ST elevation alone with no Q code. ‘‘Nonspecific ECG’’: ST segment depression and T wave inversion. ‘‘Diagnostic biomarker’’: at least two measurements of the same marker taken at least 6 hr apart, showing at least one positive value and a rising or falling pattern, in the absence of noncardiac causes of biomarker elevation. ‘‘Equivocal biomarker’’: only one measurement available, which was positive, in the absence of noncardiac causes. a If patients had died within 28 days of hospital admission for myocardial infarction, they were classified as fatal myocardial infarction.
myocardial infarction, 90.7%; 95% CI 5 88.6e92.6% [774/853]) compared to shorter stays (for all diagnoses, 31.5%; 95% CI 5 27.2e36.1% [138/438] and for myocardial infarction, 47.5%; 95% CI 5 40.7e54.3% [104/219]). The most sensitive and specific biological markers, troponins, have been widely used in Danish hospitals since the beginning of 2000. However, when stratifying data according to the time of diagnosis (before or after January 1, 2000), we detected no significant differences in positive
predictive values. Additionally, no differences were observed after stratification based on whether the hospital had specialized functions of invasive cardiology or not (data not shown). Stratification based on type of diagnosis showed higher positive predictive values for patients registered with a primary vs. a secondary diagnosis, that is, 67.1% (95% CI 5 64.6e69.5%) (956/1,425) vs. 47.0% (95% CI 5 37.6e56.5%) (54/115), respectively.
Table 3 ACS diagnoses in the Danish National Patient Registry classified based on case definitions recommended by the American Heart Association and European Society of Cardiology Discharge diagnoses in the Danish National Patient Registry Case definitions Definite myocardial infarction Probable myocardial infarction Possible myocardial infarction Unstable angina pectoris Definite fatal myocardial infarction Probable fatal myocardial infarction Possible fatal myocardial infarction Medical procedure-related events
Unstable angina pectoris, n (%) 52 8 4 51 2 0 0 5
(11.2) (1.7) (0.9) (11.1) (0.4) (0.0) (0.0) (1.1)
Myocardial infarction, n (%) 742 42 20 9 39 7 4 15
(65.6) (3.7) (1.8) (0.8) (3.5) (0.6) (0.4) (1.3)
Cardiac arrest, n (%) 6 1 3 0 4 5 2 0
(9.5) (1.6) (4.8) (0.0) (6.4) (7.9) (3.2) (0.0)
Total, n (%) 800 51 27 60 45 12 6 20
(48.4) (3.1) (1.6) (3.6) (2.7) (0.7) (0.4) (1.2)
Cases total
122 (26.5)
878 (77.7)
21 (33.3)
1,021 (61.7)
Unrecognized myocardial infarction No unstable coronary event
2 (0.4) 320 (69.4)
3 (0.3) 191 (16.9)
0 (0.0) 21 (33.3)
5 (0.3) 532 (32.2)
Noncases total
322 (69.8)
194 (17.2)
21 (33.3)
537 (32.5)
0 (0.0) 17 (3.7)
6 (0.5) 52 (4.6)
13 (20.6) 8 (12.7)
19 (1.2) 77 (4.7)
461 (100.0)
1,130 (100.0)
63 (100.0)
1,654 (100.0)
Fatal case with insufficient data Medical record missing Total
Abbreviations: ACS, acute coronary syndrome. Results are given as numbers (with percentages in brackets).
192
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
Table 4 Positive predictive values of ACS diagnoses in the Danish National Patient Registry for patients discharged from a ward Discharge diagnoses in the Danish National Patient Registry
ACS verified (positive predictive value), n (%)
95% CIs
ACS not verified, n (%)
Unstable angina pectoris Myocardial infarction Cardiac arrest
113 (42.0) 834 (92.4) 17 (53.1)
36.0e48.2 90.4e94.0 34.7e70.9
156 (58.0) 69 (7.6) 15 (46.9)
269 (100.0) 903 (100.0) 32 (100.0)
Total
964 (80.1)
77.7e82.3
240 (19.9)
1,204 (100.0)
Total, n (%)
Abbreviations: ACS, acute coronary syndrome; CI, confidence interval.
Stratification based on gender revealed a substantial difference in the positive predictive value for diagnosis of ACS. The values for men were higher than those for women for all diagnoses examined. Diagnoses of men had a positive predictive value of 72.6% (95% CI 5 69.9e75.3%) (775/ 1,067) compared to 50.1% (95% CI 5 45.6e54.6%) (246/ 491) for diagnoses of women. A subdiagnosis of myocardial infarction (adjusted odds ratio [OR] 14.8, 95% CI 5 10.6e20.5 when compared to a diagnosis of unstable angina) and a discharge diagnosis from a ward (adjusted OR 20.0, 95% CI 5 13.6e29.3 when compared to from an emergency room) were both strongly associated with having a validated diagnosis when we included the above-mentioned characteristics in a multivariable logistic regression analysis. Other characteristics associated with a validated diagnosis in this analysis were male gender and a primary diagnosis (compared to a secondary diagnosis) (Table 5).
4. Discussion We found that the positive predictive values of ACS diagnoses in the Danish National Patient Registry varied substantially for the specific subdiagnosis (myocardial infarction, unstable angina pectoris, and cardiac arrest) and depended on the type of hospital department, type of diagnosis, and gender. The highest predictive values were found for patients with myocardial infarction, those discharged from a ward, those with primary ACS diagnoses, and for male patients. The strengths of our study included the access to medical records with detailed clinical data for more than 95% of the patients in the cohort, the use of updated internationally recommended definitions of ACS, and a relatively large sample size. The latter resulted in an improved statistical precision compared to previous studies, and made it possible to perform subgroup analysis. The patients included in this study were all participants in an ongoing cohort study, with a modest participation rate, being conducted in the two largest cities in Denmark. Therefore, the patients and the hospitals in this study might not be representative of the Danish National Patient Registry as a whole. Disadvantaged population groups are often
underrepresented in cohort studies, which necessitates active participation. This appears to be the case in the ‘‘Diet, Cancer and Health’’ study [11]. However, because the Danish National Health Service provides free universal tax-supported health care for all inhabitants, there are no financial incentives for hospitals not to not provide a similar diagnostic workup for all patients, regardless of their socioeconomic position. However, health and health care disparities may persist due to reasons beyond financial incentives related to coverage of services. This can result in disparities of the sensitivity of hospital discharge registries and could at least also in theory affect the predictive value of a discharge diagnosis. Major socioeconomic disparities in the predictive value are, however, less likely to occur with conditions characterized by a thorough diagnostic workup as was the case in this study. Furthermore, the predictive values did not differ based on the type of hospital (with or without specialized functions of invasive cardiology). Therefore, we believe it is unlikely that the predictive value Table 5 List over ORs of having a validated diagnosis in the Danish National Patient Registry in different subgroups found by logistic regression analyses Subgroups
OR
95% CIs
Unstable angina pectoris Myocardial infarction Cardiac arrest
1.0 14.8 8.7
10.6e20.5 4.1e18.5
Emergency room/outpatient clinic Ward
1.0 20.0
13.6e29.3
Male Female
1.0 0.5
0.3e0.6
Agea
1.02
1.0e1.1
Hospital stay !2 days Hospital stay >2 days
1.0 1.0
1.0e1.0
No biomarker Biomarker
1.0 1.1
0.7e1.8
Primary diagnosis Secondary diagnosis
1.0 0.2
0.1e0.4
Abbreviations: CI, confidence interval; OR, odds ratio. a OR for age is calculated as a continuous variable giving the OR per 1 yr more than 50 yr of age.
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
of ACS diagnoses found in our study differs substantially from the predictive value for the general population and for other hospitals in Denmark. We used the medical records as the gold standard when assessing the predictive value of the diagnoses recorded in the National Patient Registry. Although this is the conventional approach used in most validation studies, medical records are not perfect, and it can be difficult to assess the quality of the information they contain. In our study, most of the medical records were evaluated by a single reviewer. This is a possible limitation; however, it should be remembered that more than 80% of the patients were examined with very sensitive biomarkers (troponins and CKMB), which reduced the impact of any possible subjective interpretation of symptoms and ECG recordings. It does not, however, minimize the possibilities of errors in abstracting the information from the medical records. Finally, we were not able to determine the sensitivity nor the specificity of the Danish National Patient Registry data; this would have required an additional gold standard, that is, a valid diseasespecific data source including all patients admitted with ACS during the study period. Unfortunately, such a data source is not available in Denmark. Nearly all patients who received a registered ACS diagnosis from an emergency room or outpatient clinic were immediately transferred to a ward; therefore, they might have been registered with an ACS diagnosis from both an emergency room or an outpatient clinic and a ward during the same admission. In this case, we categorized the patient as a ‘‘ward patient.’’ Patients were categorized as ‘‘emergency room patients’’ or as ‘‘outpatient clinic patients’’ if they had an ACS diagnosis registered only from the emergency room or from the outpatient clinic and were subsequently discharged from a ward with another non-ACS diagnosis. This approach partly explains the lower positive predictive value found for ACS coded at emergency rooms or outpatient clinics in our study. However, other factors may also have influenced this finding. The opportunity to examine and observe patients in emergency rooms or outpatient clinics is typically limited compared to a ward, and the stay is often short. Moreover, the diagnostic workup of emergency room patients might potentially be hampered by the severity of the disease and the impact of possible comorbidity. Therefore, it is not surprising to find a higher positive predictive value for patients categorized as ward patients compared to those categorized as emergency room or outpatient clinic patients. The diagnosis of myocardial infarction was associated with a high positive predictive value, presumably because of the stringent diagnostic criteria, including coronary markers, ECG findings, and symptoms. In contrast, the diagnosis of unstable angina pectoris resulted in a very low positive predictive value. This could be due to discrepancies between the diagnostic criteria used in the clinics and those recommended by the American Heart Association and the European Society of Cardiology for use in
193
epidemiological studies. In clinical practice, a patient with typical symptoms and even minor changes in ECG (ST depression or inverted T waves) might be diagnosed with having unstable angina pectoris, whereas they are not considered as having an acute coronary event according to the recommendations of the American Heart Association and the European Society of Cardiology. In addition, patients with minor elevations in coronary biomarkers are often diagnosed as having unstable angina pectoris in clinical practice, but in this study, these patients were classified as having myocardial infarction. Few registry studies have focused on unstable angina pectoris; however, in a study of the Finnish Hospital Discharge Register using the same definition of myocardial infarction as used here, Pajunen et al. [7] found that some events of myocardial infarction were classified as unstable angina pectoris in the registry. They also found that the positive predictive values of unstable angina pectoris and myocardial infarction when considered together were lower than the positive predictive value of myocardial infarction diagnosis alone. The validity of the diagnosis of cardiac arrest was also low. However, more than one third of the medical records of patients with cardiac arrest were either not available or did not contain sufficient data, making examination of the predictive value of this diagnosis dubious. Madsen et al. [6] compared discharge diagnoses from the Danish National Patient Registry from 1982 to 1991 with data from the DANMONICA study, medical records, discharge cards, emergency ward reports, death certificates, and autopsy reports. They found high positive predictive values and sensitivities for the myocardial infarction diagnoses registered in the Danish National Patient Registry. Our results agree with those of Madsen et al. when comparing similar data (patients with myocardial infarction discharged from a ward). In 1994, the Danish National Patient Registry was expanded to include patients from emergency rooms or from outpatient clinics in addition to patients from hospital wards; a new concept of ACS has also been implemented. Furthermore, the registration in the Danish National Patient Registry has changed from using the ICD-8 classification to the newer ICD-10 classification. This might explain the somewhat lower overall positive predictive values for all ACS diagnoses found in our study. The primary objectives for hospital discharge registries, including the Danish National Patient Registry, are administrative rather than scientific. In recent years, the Danish National Patient Registry has increasingly been used for accounting purposes, which could have influenced the classification of diagnoses in the Registry and thereby the scientific validity of the data. Moreover, there have been considerable changes in the diagnostic definition of myocardial infarction since the inception of our study. To control for these changes, we stratified the data based on the time of admission (before or after 2000). We found no difference in positive predictive values before and after 2000.
194
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
Recent studies indicate that women with ACS are more likely to experience atypical symptoms and may be more difficult to diagnose, resulting in an underestimation of the incidence of ACS among women [15,16]. Therefore, we expect the sensitivity of ACS hospital discharge diagnoses of women to be lower than that of men. However, during the course of this study (1993e2003), the definition of ACS, used both in clinics and in recommendations from the European Society of Cardiology and American Heart Association, was the same for both genders; therefore, it is somewhat surprising that we found a significantly lower positive predictive value for women than for men. This could indicate that the problems related to the registration of ACS among women are not restricted to the sensitivity of the registration systems. In conclusion, we found a high predictive value (O90%) of the diagnosis of myocardial infarction for patients discharged from a ward when we used the criteria of ACS as currently recommended by the European Society of Cardiology. These results agree with those from a previous study completed before the implementation of new diagnostic strategies for ACS [6]. However, the overall positive predictive value of ACS for all patients registered in the Danish National Patient Registry was rather low, particularly because the diagnoses of patients discharged with unstable angina pectoris and the diagnoses of patients from an emergency room or from an outpatient clinic had low positive predictive values. Our data indicate that ACS data from hospital discharge registries should be used with caution in registry studies; data validation is recommended. It might be possible to improve the predictive value of the ACS diagnoses from registries by restricting the case definition to patients with myocardial infarction, to patients discharged from wards, or to patients with a primary ACS diagnosis. Thus, our study demonstrates a feasible and simple method that may ensure higher predictive values while using ACS diagnosis data from a hospital discharge registry.
Acknowledgment The study was supported by the MD Kopps Foundation and The Research Foundation of the Danish Medical Association in North Jutland. References [1] Beaglehole R. International trends in coronary heart disease mortality and incidence rates. J Cardiovasc Risk 1999;6:63e8.
[2] Hammar N, Alfredsson L, Rosen M, Spetz CL, Kahan T, Ysberg AS. A national record linkage to study acute myocardial infarction incidence and case fatality in Sweden. Int J Epidemiol 2001;30(Suppl. 1):S30e4. [3] Kattainen A, Salomaa V, Harkanen T, Jula A, Kaaja R, Kesaniemi YA, et al. Coronary heart disease: from a disease of middle-aged men in the late 1970s to a disease of elderly women in the 2000s. Eur Heart J 2006;27:296e301. [4] Reitsma JB, Dalstra JA, Bonsel GJ, van der Meulen JH, Koster RW, Gunning-Schepers LJ, et al. Cardiovascular disease in the Netherlands, 1975 to 1995: decline in mortality, but increasing numbers of patients with chronic conditions. Heart 1999;82:52e6. [5] Mahonen M, Salomaa V, Brommels M, Molarius A, Miettinen H, Pyorala K, et al. The validity of hospital discharge register data on coronary heart disease in Finland. Eur J Epidemiol 1997;13:403e15. [6] Madsen M, Davidsen M, Rasmussen S, Abildstrom SZ, Osler M. The validity of the diagnosis of acute myocardial infarction in routine statistics: a comparison of mortality and hospital discharge data with the Danish MONICA registry. J Clin Epidemiol 2003;56:124e30. [7] Pajunen P, Koukkunen H, Ketonen M, Jerkkola T, Immonen-Raiha P, Karja-Koskenkari P, et al. The validity of the Finnish Hospital Discharge Register and Causes of Death Register data on coronary heart disease. Eur J Cardiovasc Prev Rehabil 2005;12:132e7. [8] Pladevall M, Goff DC, Nichaman MZ, Chan F, Ramsey D, Ortiz C, et al. An assessment of the validity of ICD Code 410 to identify hospital admissions for myocardial infarction: the Corpus Christi Heart Project. Int J Epidemiol 1996;25:948e52. [9] Boyle CA, Dobson AJ. The accuracy of hospital records and death certificates for acute myocardial infarction. Aust NZ J Med 1995;25:316e23. [10] Myocardial infarction redefinedda consensus document of the Joint European Society of Cardiology/American College of Cardiology Committee for the redefinition of myocardial infarction. Eur Heart J 2000;21:1502e13. [11] Tjonneland A, Olsen A, Boll K, Stripp C, Christensen J, Engholm G, et al. Study design, exposure variables, and socioeconomic determinants of participation in Diet, Cancer and Health: a population-based prospective cohort study of 57,053 men and women in Denmark. Scand J Pub Health 2007;35:432e41. [12] Andersen TF, Madsen M, Jorgensen J, Mellemkjoer L, Olsen JH. The Danish National Hospital Register. A valuable source of data for modern health sciences. Dan Med Bull 1999;46:263e8. [13] Luepker RV, Apple FS, Christenson RH, Crow RS, Fortmann SP, Goff D, et al. Case definitions for acute coronary heart disease in epidemiology and clinical research studies: a statement from the AHA Council on Epidemiology and Prevention; AHA Statistics Committee; World Heart Federation Council on Epidemiology and Prevention; the European Society of Cardiology Working Group on Epidemiology and Prevention; Centers for Disease Control and Prevention; and the National Heart, Lung, and Blood Institute. Circulation 2003;108:2543e9. [14] Pedersen CB, Gotzsche H, Moller JO, Mortensen PB. The Danish Civil Registration System. A cohort of eight million persons. Dan Med Bull 2006;53:441e9. [15] Lovlien M, Schei B, Hole T. Women with myocardial infarction are less likely than men to experience chest symptoms. Scand Cardiovasc J 2006;40:342e7. [16] Elsaesser A, Hamm CW. Acute coronary syndrome: the risk of being female. Circulation 2004;109:565e7.
Predictive values of acute coronary syndrome discharge diagnoses differed in the Danish National Patient Registry Albert Marni Joensena,*, Majken K. Jensenb, Kim Overvadb, Claus Dethlefsenc, Erik Schmidta, Lars Rasmussena, Anne Tjønnelandd, Søren Johnsenb a
Department of Cardiology, Center for Cardiovascular Research, Aalborg Hospital, Aarhus University Hospital, Sdr. Skovvej 15, DK-9000 Aalborg, Denmark b Department of Clinical Epidemiology, Center for Cardiovascular Research, Aarhus University Hospital, Aalborg, Denmark c Center for Cardiovascular Research, Aalborg Hospital, Aarhus University Hospital, Aalborg, Denmark d Danish Cancer Society, Institute of Epidemiology, Copenhagen, Denmark Accepted 31 March 2008
Abstract Objective: To investigate the predictive value of acute coronary syndrome (ACS) diagnoses, including unstable angina pectoris, myocardial infarction, and cardiac arrest, in the Danish National Patient Registry. Study Design and Setting: We identified all first-time ACS diagnoses in the Danish National Patient Registry among participants in the Danish cohort study ‘‘Diet, Cancer and Health’’ through the end of 2003. We retrieved and reviewed medical records based on current European Society of Cardiology criteria for ACS. Results: We reviewed hospital medical records of 1,577 out of 1,654 patients (95.3%) who had been hospitalized with a first-time ACS diagnosis. The overall positive predictive value for ACS was 65.5% (95% confidence interval [CI] 5 63.1e67.9%). Stratification by subdiagnosis and hospital department produced significantly higher positive predictive values for myocardial infarction diagnoses (81.9%; 95% CI 5 79.5e84.2%) and among patients who received an ACS diagnosis in a ward (80.1%; 95% CI 5 77.7e82.3%). Conclusion: The ACS diagnoses contained in hospital discharge registries should be used with caution. If validation is not possible, restricting analyses to patients with myocardial infarction and/or patients discharged from wards might be a useful alternative. Ó 2008 Elsevier Inc. All rights reserved. Keywords: Validation studies; Coronary disease; Myocardial infarction; Registries; Unstable angina pectoris; Diagnosis
1. Introduction The mortality from coronary heart disease (CHD) has declined over recent decades in most industrialized countries; however, CHD remains a leading cause of death and morbidity [1e4]. Major efforts have been made by physicians, administrators, and politicians to further reduce the risk and further improve the prognosis of CHD. Therefore, it is essential that up-to-date and valid data on the incidence and outcome of CHD are available. Hospital discharge and other disease registries are cost-efficient data sources; however, their usability is highly dependent on the validity of the registered data. Previous studies on the validity of CHD diagnoses contained in hospital discharge and other disease registries
* Corresponding author. Tel.: þ45-99-32-68-39. E-mail address: [email protected] (A.M. Joensen). 0895-4356/09/$ e see front matter Ó 2008 Elsevier Inc. All rights reserved. doi: 10.1016/j.jclinepi.2008.03.005
have focused primarily on myocardial infarction. In general, these studies have reported sensitivities with some variation in the positive predictive value (65e96%) [2,5e9]. Few studies have examined the validity of unstable angina pectoris and cardiac arrest diagnoses, and most of these studies were performed before the introduction of the International Classification of Diseases’ 10th revision (ICD-10). Moreover, the strategy for diagnosing patients with suspected acute coronary syndrome (ACS), that is, myocardial infarction, unstable angina pectoris, and cardiac arrest, has changed considerably during the last few years due to the introduction of sensitive and specific biomarkers of myocardial necrosis and a new definition of myocardial infarction that includes troponin levels [10]. These changes would presumably have implications for the diagnostic workup and validity of the ACS diagnosis. Therefore, we examined the predictive value of myocardial infarction, unstable angina pectoris, and cardiac arrest diagnoses contained in the Danish National Patient Registry for
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
participants in the cohort study ‘‘Diet, Cancer and Health’’ from 1994 to 2003.
2. Methods 2.1. ‘‘Diet, Cancer and Health’’ ‘‘Diet, Cancer and Health’’ is a prospective cohort study with the primary objective of analyzing the etiological role of diet in the development of cancer. The study has been described in detail elsewhere [11]. Between December 1993 and May 1997, 80,996 men and 79,729 women aged 50e64 years were invited to participate in the study; 27,179 men and 29,876 women accepted the invitation. Those born in Denmark, living in the urban areas of Copenhagen and Aarhus, and not registered with a cancer diagnosis in the Danish Cancer Registry at the time of invitation were eligible cohort members.
189
Patient Registry, including primary and secondary diagnoses. Participants who had an ACS discharge diagnosis before enrollment in ‘‘Diet, Cancer and Health’’ were excluded. Only participants who had received a first-time ACS diagnosis in the Danish National Patient Registry after the time of their enrollment in ‘‘Diet, Cancer and Health’’ until January 1, 2004 were included in the study. We noted whether the department of admission was an emergency room, an outpatient clinic, or a ward, and whether the hospital of admission had specialized functions of invasive cardiology or not. If patients who visited an emergency room were admitted to a hospital ward and were registered with an ACS diagnosis by both departments, they were categorized as patients from a ward. When registered with an ACS diagnosis from an emergency room (or an outpatient clinic) without a subsequent ACS diagnosis from a ward, they were categorized as patients from an emergency room (or an outpatient clinic).
2.2. Identification of possible cases of ACS The Danish National Health Service provides free universal tax-supported health care, including hospital care, for all inhabitants (Health Care in Denmark. Copenhagen: Ministry of the Interior and Health, 2003; available at: http://www.im.dk/publikationer/healthcare_in_dk/index.htm). Patients with acute medical conditions are admitted to and treated in public hospitals. The Danish National Patient Registry was established in 1977, and 99.4% of all discharges from somatic hospitals (including public and private hospitals) have been recorded in this database [12]. If a patient was transferred from one department to another, the patient was registered with a discharge diagnosis from the first department at the time of transfer and subsequently registered with a second diagnosis when discharged from the second department. The diagnoses from the two departments need not be identical. Since 1995, discharges from emergency rooms and outpatient clinics have also been included in the registry. The registered data from each admission include the civil registry number, which is the unique personal identification number of every Danish citizen, the dates of admission and discharge, the surgical procedures performed, and up to 20 discharge diagnoses classified according to the Danish version of the International Classification of Diseases, 8th revision (ICD-8) until 1993. Since 1994, diagnoses have been classified according to the corresponding national version of ICD-10 [12]. All discharge diagnoses are determined exclusively by the physician who discharges the patient and cannot be altered later, for example, for administrative or financial reason. Based on the available hospital discharge history of each participant, we identified participants in the ‘‘Diet, Cancer and Health’’ cohort who were registered with a discharge diagnosis of ACS (ICD-8 410e410.99, 427.27 and ICD10 I20.0, I21.0eI21.9, I46.0eI46.9) in the Danish National
2.3. Medical record review We retrieved medical records corresponding to the discharge diagnosis and exact discharge date contained in the National Patient Registry from 54 different hospitals. We sought to retrieve the medical records that covered the hospital admissions recorded in the National Registry of Patients; however, minor mismatches (eg, 1 or 2 days in admission or discharge date) between the registry and the medical records were accepted if all other available information indicated that the medical record covered the admission identified in the registry. The records were reviewed by one of three reviewers (AMJ, SPJ, or MKJ). If the complete medical record was not available, discharge letters, results of blood tests, and electrocardiograms (ECG) were retrieved whenever possible. Information about clinical symptoms, ECGs, and biomarkers of myocardial infarction was noted. If the patient had died within 28 days of developing a myocardial infarction, the date of death was noted and the autopsy record, if any, was retrieved. We also noted the duration of the hospital stay. Patients were classified in accordance with the current recommendations of the American Heart Association and the European Society of Cardiology, as described by Luepker et al. [13]. This classification is based on symptoms, signs, coronary biomarkers (creatine kinase, creatine kinase B, creatine kinase (CK) MB, troponin T, troponin I, lactate dehydrogenase, and lactate dehydrogenase 1), and ECG and/or autopsy findings. All cases with an uncertain diagnosis based on the available information were discussed, and diagnoses were made according to consensus. We used the Civil Registration System, which retains data on vital status, address, and emigration for the Danish population since 1968, to obtain information on patient deaths [14].
190
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
For more details on the classification of ACS events, please see Luepker et al. [13]. 2.4. Statistical analyses Patients with definite myocardial infarction (fatal and nonfatal events), probable myocardial infarction, possible myocardial infarction, unstable angina pectoris, or a medical procedure-related event were categorized as verified ACS. Patients with no unstable coronary events or unrecognized myocardial infarction were categorized as verified non-ACS. Patients for whom information was insufficient to categorize them as either verified ACS or verified nonACS were excluded from the analyses. Positive predictive values of ACS diagnoses registered in the Danish National Patient Registry were calculated as proportions, that is, the numerator contained the number of patients with verified ACS after review of medical records and the denominator contained the total number of patients registered in the Danish National Patient Registry with this specific diagnosis. Positive predictive values were calculated for ACS as a whole and for each specific ACS subdiagnosis (unstable angina pectoris, myocardial infarction, cardiac arrest). Furthermore, data were stratified based on the type of department (ward, emergency room, or outpatient clinic), length of hospital stay, time of diagnosis (before or after January 1, 2000), type of hospital (whether the hospital of admission had specialized functions of invasive cardiology or not), and gender. Finally, we conducted a multivariable logistic regression to estimate the risk of having a validated diagnosis in order to identify characteristics independently associated with the positive predictive value. In this analysis, we included the following factors: subdiagnosis, type of department, gender, age, length of stay, type of biomarkers (troponins/ CKMB, others, none), and type of diagnosis (primary or secondary). All the factors were mutually adjusted.
3. Results We identified 1,654 patients with an incident diagnosis of ACS in the Danish National Patient Registry among the participants in the ‘‘Diet, Cancer and Health’’ study. Baseline characteristics of the cohort are presented in Table 1. The cohort provided 419,949 person-years of risk resulting in an incidence rate of first-time ACS diagnosis in the Danish National Patient Registry of 3.9 events per 1,000 person-years (95% confidence interval [CI] 5 3.7e4.1). We were able to retrieve medical records or discharge letters for 1,577 (95.3%) of the patients. We were not able to characterize 96 patients, either because we could not retrieve their medical records (n 5 77) or because their medical records had insufficient data to classify these patients (n 5 19). From the medical records, we were able to find information on symptoms (n 5 1546; 98.0%), ECG
Table 1 Baseline characteristics of the ‘‘Diet, Cancer and Health Study’’ population Characteristics
Study population
Age at enrollmenta (yr) Current smokers (%) Body mass index O25 kg/m2 (%) Total cholesterola (mmol/L) Blood pressurea (mmHg) Diabetes mellitus (%) Education !8 yr (%)
56.1 (52.4e60.3) 36.1 55.9 6.0 (5.3e6.8) 138 (125e152) 2.0 32.8
a
Median (1. quartile and 3. quartile in brackets).
(n 5 1,501; 95.2%), all biomarkers (n 5 1,363; 86.4%), and highly specific biomarkers, troponins, and CKMB (n 5 1,290; 81.8%). Table 2 shows how patients are classified in accordance with the American Heart Association and European Society of Cardiology guidelines. Classification of the ACS diagnoses is presented in Table 3. The positive predictive value for ACS as a whole after exclusion of those with missing medical records was 65.5% (95% CI 5 63.1e67.9%) (1,021/1,558). After stratification by subdiagnoses, we found a substantially higher overall positive predictive value for myocardial infarction (81.9%; 95% CI 5 79.5e84.2%) (878/1,072) compared to unstable angina pectoris (27.5%; 95% CI 5 23.4e31.9%) (122/444) and cardiac arrest (50.0%; 95% CI 5 34.2e65.8%) (21/42). When we stratified the data based on the type of department of discharge, the predictive value for patients diagnosed in a ward was much higher (80.1%; 95% CI 5 77.7e82.3%) (964/1,204) than that for patients diagnosed in an emergency room or in an outpatient clinic (16.1%; 95% CI 5 12.4e20.4%) (57/354). There were 1,021 patients with a verified diagnosis of ACS, 834 (81.7%) of whom were discharged from a ward with a diagnosis of myocardial infarction. Only 57 (5.6%) of the patients with verified ACS were discharged from an emergency room or from an outpatient clinic (among these, 10 patients died). Nearly all the patients registered with an ACS diagnosis from an emergency room or outpatient clinic were transferred to a hospital ward and subsequently discharged with diagnoses other than ACS. These patients often had stable angina pectoris or had chest pain for other reasons. The positive predictive value was as high as 92.4% (95% CI 5 90.4e94.0%) (834/903) for patients diagnosed with a myocardial infarction in a ward compared to 26.0% (95% CI 5 19.6e33.3%) (44/169) for patients diagnosed with a myocardial infarction in an emergency room or in an outpatient clinic (Table 4). Stratification based on the length of hospital stay (longer vs. 2 days or less) produced similar but not so pronounced differences in positive predictive values for patients with hospital stays longer than 2 days (for all diagnoses, 78.8%; 95% CI 5 76.3e81.2% [883/1,120] and for
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
191
Table 2 Classification of cardiac events based on symptoms/signs, ECGs, and/or biomarkers, as recommended by the American Heart Association and European Society of Cardiology Definite myocardial infarctiona
Diagnostic biomarker findings or evolving ECG Or, if fatal, typical autopsy findings Equivocal biomarker findings and positive ECG If biomarkers are missing, positive ECG and typical symptoms Equivocal biomarker findings and (nonspecific ECG or typical symptoms) If biomarkers are missing, positive ECG Normal biomarkers and positive ECG and typical symptoms Cardiac event (as described above) within a period of 28 days after a medical procedure (eg, general surgery) Appearance, in a nonacute setting, of a new diagnostic Q wave
Probably myocardial infarctiona Possible myocardial infarctiona Unstable angina pectoris Medical procedure-related event Unrecognized myocardial infarction
Abbreviations: ECG, electrocardiograms; CHD, coronary heart disease. Symptoms or signs were categorized as either typical (acute chest, epigastric, neck, jaw or arm pain, discomfort or pressure without any apparent noncardiac source, or acute congestive heart failure in the absence of non-CHD causes) or atypical. ‘‘Evolving ECG’’: evolvement of a Q code. ‘‘Positive ECG’’: evolving ST elevation alone with no Q code. ‘‘Nonspecific ECG’’: ST segment depression and T wave inversion. ‘‘Diagnostic biomarker’’: at least two measurements of the same marker taken at least 6 hr apart, showing at least one positive value and a rising or falling pattern, in the absence of noncardiac causes of biomarker elevation. ‘‘Equivocal biomarker’’: only one measurement available, which was positive, in the absence of noncardiac causes. a If patients had died within 28 days of hospital admission for myocardial infarction, they were classified as fatal myocardial infarction.
myocardial infarction, 90.7%; 95% CI 5 88.6e92.6% [774/853]) compared to shorter stays (for all diagnoses, 31.5%; 95% CI 5 27.2e36.1% [138/438] and for myocardial infarction, 47.5%; 95% CI 5 40.7e54.3% [104/219]). The most sensitive and specific biological markers, troponins, have been widely used in Danish hospitals since the beginning of 2000. However, when stratifying data according to the time of diagnosis (before or after January 1, 2000), we detected no significant differences in positive
predictive values. Additionally, no differences were observed after stratification based on whether the hospital had specialized functions of invasive cardiology or not (data not shown). Stratification based on type of diagnosis showed higher positive predictive values for patients registered with a primary vs. a secondary diagnosis, that is, 67.1% (95% CI 5 64.6e69.5%) (956/1,425) vs. 47.0% (95% CI 5 37.6e56.5%) (54/115), respectively.
Table 3 ACS diagnoses in the Danish National Patient Registry classified based on case definitions recommended by the American Heart Association and European Society of Cardiology Discharge diagnoses in the Danish National Patient Registry Case definitions Definite myocardial infarction Probable myocardial infarction Possible myocardial infarction Unstable angina pectoris Definite fatal myocardial infarction Probable fatal myocardial infarction Possible fatal myocardial infarction Medical procedure-related events
Unstable angina pectoris, n (%) 52 8 4 51 2 0 0 5
(11.2) (1.7) (0.9) (11.1) (0.4) (0.0) (0.0) (1.1)
Myocardial infarction, n (%) 742 42 20 9 39 7 4 15
(65.6) (3.7) (1.8) (0.8) (3.5) (0.6) (0.4) (1.3)
Cardiac arrest, n (%) 6 1 3 0 4 5 2 0
(9.5) (1.6) (4.8) (0.0) (6.4) (7.9) (3.2) (0.0)
Total, n (%) 800 51 27 60 45 12 6 20
(48.4) (3.1) (1.6) (3.6) (2.7) (0.7) (0.4) (1.2)
Cases total
122 (26.5)
878 (77.7)
21 (33.3)
1,021 (61.7)
Unrecognized myocardial infarction No unstable coronary event
2 (0.4) 320 (69.4)
3 (0.3) 191 (16.9)
0 (0.0) 21 (33.3)
5 (0.3) 532 (32.2)
Noncases total
322 (69.8)
194 (17.2)
21 (33.3)
537 (32.5)
0 (0.0) 17 (3.7)
6 (0.5) 52 (4.6)
13 (20.6) 8 (12.7)
19 (1.2) 77 (4.7)
461 (100.0)
1,130 (100.0)
63 (100.0)
1,654 (100.0)
Fatal case with insufficient data Medical record missing Total
Abbreviations: ACS, acute coronary syndrome. Results are given as numbers (with percentages in brackets).
192
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
Table 4 Positive predictive values of ACS diagnoses in the Danish National Patient Registry for patients discharged from a ward Discharge diagnoses in the Danish National Patient Registry
ACS verified (positive predictive value), n (%)
95% CIs
ACS not verified, n (%)
Unstable angina pectoris Myocardial infarction Cardiac arrest
113 (42.0) 834 (92.4) 17 (53.1)
36.0e48.2 90.4e94.0 34.7e70.9
156 (58.0) 69 (7.6) 15 (46.9)
269 (100.0) 903 (100.0) 32 (100.0)
Total
964 (80.1)
77.7e82.3
240 (19.9)
1,204 (100.0)
Total, n (%)
Abbreviations: ACS, acute coronary syndrome; CI, confidence interval.
Stratification based on gender revealed a substantial difference in the positive predictive value for diagnosis of ACS. The values for men were higher than those for women for all diagnoses examined. Diagnoses of men had a positive predictive value of 72.6% (95% CI 5 69.9e75.3%) (775/ 1,067) compared to 50.1% (95% CI 5 45.6e54.6%) (246/ 491) for diagnoses of women. A subdiagnosis of myocardial infarction (adjusted odds ratio [OR] 14.8, 95% CI 5 10.6e20.5 when compared to a diagnosis of unstable angina) and a discharge diagnosis from a ward (adjusted OR 20.0, 95% CI 5 13.6e29.3 when compared to from an emergency room) were both strongly associated with having a validated diagnosis when we included the above-mentioned characteristics in a multivariable logistic regression analysis. Other characteristics associated with a validated diagnosis in this analysis were male gender and a primary diagnosis (compared to a secondary diagnosis) (Table 5).
4. Discussion We found that the positive predictive values of ACS diagnoses in the Danish National Patient Registry varied substantially for the specific subdiagnosis (myocardial infarction, unstable angina pectoris, and cardiac arrest) and depended on the type of hospital department, type of diagnosis, and gender. The highest predictive values were found for patients with myocardial infarction, those discharged from a ward, those with primary ACS diagnoses, and for male patients. The strengths of our study included the access to medical records with detailed clinical data for more than 95% of the patients in the cohort, the use of updated internationally recommended definitions of ACS, and a relatively large sample size. The latter resulted in an improved statistical precision compared to previous studies, and made it possible to perform subgroup analysis. The patients included in this study were all participants in an ongoing cohort study, with a modest participation rate, being conducted in the two largest cities in Denmark. Therefore, the patients and the hospitals in this study might not be representative of the Danish National Patient Registry as a whole. Disadvantaged population groups are often
underrepresented in cohort studies, which necessitates active participation. This appears to be the case in the ‘‘Diet, Cancer and Health’’ study [11]. However, because the Danish National Health Service provides free universal tax-supported health care for all inhabitants, there are no financial incentives for hospitals not to not provide a similar diagnostic workup for all patients, regardless of their socioeconomic position. However, health and health care disparities may persist due to reasons beyond financial incentives related to coverage of services. This can result in disparities of the sensitivity of hospital discharge registries and could at least also in theory affect the predictive value of a discharge diagnosis. Major socioeconomic disparities in the predictive value are, however, less likely to occur with conditions characterized by a thorough diagnostic workup as was the case in this study. Furthermore, the predictive values did not differ based on the type of hospital (with or without specialized functions of invasive cardiology). Therefore, we believe it is unlikely that the predictive value Table 5 List over ORs of having a validated diagnosis in the Danish National Patient Registry in different subgroups found by logistic regression analyses Subgroups
OR
95% CIs
Unstable angina pectoris Myocardial infarction Cardiac arrest
1.0 14.8 8.7
10.6e20.5 4.1e18.5
Emergency room/outpatient clinic Ward
1.0 20.0
13.6e29.3
Male Female
1.0 0.5
0.3e0.6
Agea
1.02
1.0e1.1
Hospital stay !2 days Hospital stay >2 days
1.0 1.0
1.0e1.0
No biomarker Biomarker
1.0 1.1
0.7e1.8
Primary diagnosis Secondary diagnosis
1.0 0.2
0.1e0.4
Abbreviations: CI, confidence interval; OR, odds ratio. a OR for age is calculated as a continuous variable giving the OR per 1 yr more than 50 yr of age.
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
of ACS diagnoses found in our study differs substantially from the predictive value for the general population and for other hospitals in Denmark. We used the medical records as the gold standard when assessing the predictive value of the diagnoses recorded in the National Patient Registry. Although this is the conventional approach used in most validation studies, medical records are not perfect, and it can be difficult to assess the quality of the information they contain. In our study, most of the medical records were evaluated by a single reviewer. This is a possible limitation; however, it should be remembered that more than 80% of the patients were examined with very sensitive biomarkers (troponins and CKMB), which reduced the impact of any possible subjective interpretation of symptoms and ECG recordings. It does not, however, minimize the possibilities of errors in abstracting the information from the medical records. Finally, we were not able to determine the sensitivity nor the specificity of the Danish National Patient Registry data; this would have required an additional gold standard, that is, a valid diseasespecific data source including all patients admitted with ACS during the study period. Unfortunately, such a data source is not available in Denmark. Nearly all patients who received a registered ACS diagnosis from an emergency room or outpatient clinic were immediately transferred to a ward; therefore, they might have been registered with an ACS diagnosis from both an emergency room or an outpatient clinic and a ward during the same admission. In this case, we categorized the patient as a ‘‘ward patient.’’ Patients were categorized as ‘‘emergency room patients’’ or as ‘‘outpatient clinic patients’’ if they had an ACS diagnosis registered only from the emergency room or from the outpatient clinic and were subsequently discharged from a ward with another non-ACS diagnosis. This approach partly explains the lower positive predictive value found for ACS coded at emergency rooms or outpatient clinics in our study. However, other factors may also have influenced this finding. The opportunity to examine and observe patients in emergency rooms or outpatient clinics is typically limited compared to a ward, and the stay is often short. Moreover, the diagnostic workup of emergency room patients might potentially be hampered by the severity of the disease and the impact of possible comorbidity. Therefore, it is not surprising to find a higher positive predictive value for patients categorized as ward patients compared to those categorized as emergency room or outpatient clinic patients. The diagnosis of myocardial infarction was associated with a high positive predictive value, presumably because of the stringent diagnostic criteria, including coronary markers, ECG findings, and symptoms. In contrast, the diagnosis of unstable angina pectoris resulted in a very low positive predictive value. This could be due to discrepancies between the diagnostic criteria used in the clinics and those recommended by the American Heart Association and the European Society of Cardiology for use in
193
epidemiological studies. In clinical practice, a patient with typical symptoms and even minor changes in ECG (ST depression or inverted T waves) might be diagnosed with having unstable angina pectoris, whereas they are not considered as having an acute coronary event according to the recommendations of the American Heart Association and the European Society of Cardiology. In addition, patients with minor elevations in coronary biomarkers are often diagnosed as having unstable angina pectoris in clinical practice, but in this study, these patients were classified as having myocardial infarction. Few registry studies have focused on unstable angina pectoris; however, in a study of the Finnish Hospital Discharge Register using the same definition of myocardial infarction as used here, Pajunen et al. [7] found that some events of myocardial infarction were classified as unstable angina pectoris in the registry. They also found that the positive predictive values of unstable angina pectoris and myocardial infarction when considered together were lower than the positive predictive value of myocardial infarction diagnosis alone. The validity of the diagnosis of cardiac arrest was also low. However, more than one third of the medical records of patients with cardiac arrest were either not available or did not contain sufficient data, making examination of the predictive value of this diagnosis dubious. Madsen et al. [6] compared discharge diagnoses from the Danish National Patient Registry from 1982 to 1991 with data from the DANMONICA study, medical records, discharge cards, emergency ward reports, death certificates, and autopsy reports. They found high positive predictive values and sensitivities for the myocardial infarction diagnoses registered in the Danish National Patient Registry. Our results agree with those of Madsen et al. when comparing similar data (patients with myocardial infarction discharged from a ward). In 1994, the Danish National Patient Registry was expanded to include patients from emergency rooms or from outpatient clinics in addition to patients from hospital wards; a new concept of ACS has also been implemented. Furthermore, the registration in the Danish National Patient Registry has changed from using the ICD-8 classification to the newer ICD-10 classification. This might explain the somewhat lower overall positive predictive values for all ACS diagnoses found in our study. The primary objectives for hospital discharge registries, including the Danish National Patient Registry, are administrative rather than scientific. In recent years, the Danish National Patient Registry has increasingly been used for accounting purposes, which could have influenced the classification of diagnoses in the Registry and thereby the scientific validity of the data. Moreover, there have been considerable changes in the diagnostic definition of myocardial infarction since the inception of our study. To control for these changes, we stratified the data based on the time of admission (before or after 2000). We found no difference in positive predictive values before and after 2000.
194
A.M. Joensen et al. / Journal of Clinical Epidemiology 62 (2009) 188e194
Recent studies indicate that women with ACS are more likely to experience atypical symptoms and may be more difficult to diagnose, resulting in an underestimation of the incidence of ACS among women [15,16]. Therefore, we expect the sensitivity of ACS hospital discharge diagnoses of women to be lower than that of men. However, during the course of this study (1993e2003), the definition of ACS, used both in clinics and in recommendations from the European Society of Cardiology and American Heart Association, was the same for both genders; therefore, it is somewhat surprising that we found a significantly lower positive predictive value for women than for men. This could indicate that the problems related to the registration of ACS among women are not restricted to the sensitivity of the registration systems. In conclusion, we found a high predictive value (O90%) of the diagnosis of myocardial infarction for patients discharged from a ward when we used the criteria of ACS as currently recommended by the European Society of Cardiology. These results agree with those from a previous study completed before the implementation of new diagnostic strategies for ACS [6]. However, the overall positive predictive value of ACS for all patients registered in the Danish National Patient Registry was rather low, particularly because the diagnoses of patients discharged with unstable angina pectoris and the diagnoses of patients from an emergency room or from an outpatient clinic had low positive predictive values. Our data indicate that ACS data from hospital discharge registries should be used with caution in registry studies; data validation is recommended. It might be possible to improve the predictive value of the ACS diagnoses from registries by restricting the case definition to patients with myocardial infarction, to patients discharged from wards, or to patients with a primary ACS diagnosis. Thus, our study demonstrates a feasible and simple method that may ensure higher predictive values while using ACS diagnosis data from a hospital discharge registry.
Acknowledgment The study was supported by the MD Kopps Foundation and The Research Foundation of the Danish Medical Association in North Jutland. References [1] Beaglehole R. International trends in coronary heart disease mortality and incidence rates. J Cardiovasc Risk 1999;6:63e8.
[2] Hammar N, Alfredsson L, Rosen M, Spetz CL, Kahan T, Ysberg AS. A national record linkage to study acute myocardial infarction incidence and case fatality in Sweden. Int J Epidemiol 2001;30(Suppl. 1):S30e4. [3] Kattainen A, Salomaa V, Harkanen T, Jula A, Kaaja R, Kesaniemi YA, et al. Coronary heart disease: from a disease of middle-aged men in the late 1970s to a disease of elderly women in the 2000s. Eur Heart J 2006;27:296e301. [4] Reitsma JB, Dalstra JA, Bonsel GJ, van der Meulen JH, Koster RW, Gunning-Schepers LJ, et al. Cardiovascular disease in the Netherlands, 1975 to 1995: decline in mortality, but increasing numbers of patients with chronic conditions. Heart 1999;82:52e6. [5] Mahonen M, Salomaa V, Brommels M, Molarius A, Miettinen H, Pyorala K, et al. The validity of hospital discharge register data on coronary heart disease in Finland. Eur J Epidemiol 1997;13:403e15. [6] Madsen M, Davidsen M, Rasmussen S, Abildstrom SZ, Osler M. The validity of the diagnosis of acute myocardial infarction in routine statistics: a comparison of mortality and hospital discharge data with the Danish MONICA registry. J Clin Epidemiol 2003;56:124e30. [7] Pajunen P, Koukkunen H, Ketonen M, Jerkkola T, Immonen-Raiha P, Karja-Koskenkari P, et al. The validity of the Finnish Hospital Discharge Register and Causes of Death Register data on coronary heart disease. Eur J Cardiovasc Prev Rehabil 2005;12:132e7. [8] Pladevall M, Goff DC, Nichaman MZ, Chan F, Ramsey D, Ortiz C, et al. An assessment of the validity of ICD Code 410 to identify hospital admissions for myocardial infarction: the Corpus Christi Heart Project. Int J Epidemiol 1996;25:948e52. [9] Boyle CA, Dobson AJ. The accuracy of hospital records and death certificates for acute myocardial infarction. Aust NZ J Med 1995;25:316e23. [10] Myocardial infarction redefinedda consensus document of the Joint European Society of Cardiology/American College of Cardiology Committee for the redefinition of myocardial infarction. Eur Heart J 2000;21:1502e13. [11] Tjonneland A, Olsen A, Boll K, Stripp C, Christensen J, Engholm G, et al. Study design, exposure variables, and socioeconomic determinants of participation in Diet, Cancer and Health: a population-based prospective cohort study of 57,053 men and women in Denmark. Scand J Pub Health 2007;35:432e41. [12] Andersen TF, Madsen M, Jorgensen J, Mellemkjoer L, Olsen JH. The Danish National Hospital Register. A valuable source of data for modern health sciences. Dan Med Bull 1999;46:263e8. [13] Luepker RV, Apple FS, Christenson RH, Crow RS, Fortmann SP, Goff D, et al. Case definitions for acute coronary heart disease in epidemiology and clinical research studies: a statement from the AHA Council on Epidemiology and Prevention; AHA Statistics Committee; World Heart Federation Council on Epidemiology and Prevention; the European Society of Cardiology Working Group on Epidemiology and Prevention; Centers for Disease Control and Prevention; and the National Heart, Lung, and Blood Institute. Circulation 2003;108:2543e9. [14] Pedersen CB, Gotzsche H, Moller JO, Mortensen PB. The Danish Civil Registration System. A cohort of eight million persons. Dan Med Bull 2006;53:441e9. [15] Lovlien M, Schei B, Hole T. Women with myocardial infarction are less likely than men to experience chest symptoms. Scand Cardiovasc J 2006;40:342e7. [16] Elsaesser A, Hamm CW. Acute coronary syndrome: the risk of being female. Circulation 2004;109:565e7.